Repeated citations are consolidated below. Chapter numbers show where each source is relied on. All sources checked 9 September 2026 unless another date is stated.
OpenAI
1. GPT-6 Astra model documentation. Context window, maximum output, pricing, reasoning levels, supported tools. developers.openai.com/api/docs/models/gpt-6-astra. Chapters 1, 2, 3, 6, 15; back matter.
2. GPT-6 Astra model guidance. Reasoning configuration, asynchronous tool calling, mid-turn steering, persisted reasoning, compaction, multi-agent orchestration, subagent delegation, misalignment monitoring. developers.openai.com/api/docs/guides/latest-model. Chapters 1, 2, 5, 6, 9, 10, 11, 15.
3. GPT-6 Astra: A new generation of intelligence. Release post covering positioning across computer use, coding, cybersecurity, science and professional work, computer-use performance, prompt-injection robustness. openai.com/index/gpt-6-astra/. Published 3 September 2026. Chapters 3, 4, 7, 15, 16.
4. Safety overview: GPT-6 Astra. Critical cybersecurity classification, trajectory monitoring, robustness, alignment and prompt-injection findings, reduced chain-of-thought monitorability. openai.com/index/safety-overview-gpt-6-astra/. Published 3 September 2026. Chapters 5, 12, 14, 15.
5. GPT-6 Astra System Card. Deployment simulation, misalignment monitoring, monitoring scopes, chain-of-thought controllability, action-only monitoring, no-CoT capability, external UK AISI findings. deploymentsafety.openai.com/gpt-6-astra. Published 3 September 2026. Chapters 12, 13, 14; back matter.
6. Research acceleration: The view inside OpenAI. Agent-workday usage, concurrent agents, experiment activity, changing research bottlenecks. openai.com/index/research-acceleration-view-inside-openai/. Published 6 September 2026. Chapter 16.
7. The Work Now Within Reach. Falling cost of intelligence and work previously constrained by time, cost or access to expertise. openai.com/index/the-work-now-within-reach/. Published 8 September 2026. Chapter 16.
Anthropic
8. Claude Fable 5.1 model documentation and effort guide. Context window, maximum output, pricing and cache rates, adaptive thinking, five-level effort parameter, per-message effort change. platform.claude.com/docs/en/models/fable-5-1/overview and platform.claude.com/docs/en/build-with-claude/effort. Chapters 2, 15; back matter.
9. Claude Fable 5.1 product page. Positioning, long-running work, tool use, failure recovery, coding and vision, self-written tests, pricing and cache-read cost, safeguard fallback behavior, research capability. anthropic.com/claude/fable. Announced 1 September 2026. Chapters 1, 2, 3, 4, 5, 9, 12, 13, 15; back matter.
10. Claude Mythos 5.1 product page. Statement that Fable 5.1 and Mythos 5.1 share the same underlying model and differ in safeguard and access treatment. anthropic.com/claude/mythos. Chapters 5, 12, 15, 16.
11. Fable cyber safeguards and jailbreak framework. Safety classifiers around Fable for sensitive use. anthropic.com/news/fable-safeguards-jailbreak-framework. Published 2 July 2026. Chapters 5, 12.
12. Claude Science: AI workbench. Connected research tools and auditable artifacts. anthropic.com/news/claude-science-ai-workbench. Chapter 4.
13. Previewing the Model Hardware Standard. AI agents operating laboratory and manufacturing instruments, parallel workflows, real-time parameter adjustment, hardware-error recovery. anthropic.com/news/model-hardware-standard-research-preview. Published 27 August 2026. Chapters 8, 16.
14. Building Multi-Agent Systems That Actually Work. Session description, Anthropic at Google Cloud Next 2026, identifying context isolation, parallel execution and specialization as recurring cases where multi-agent systems win. anthropic.com/events/anthropic-at-google-cloud-next-2026. Chapter 11.
15. Model system cards index. Listing of the September 2026 Claude Fable 5.1 and Mythos 5.1 system card. anthropic.com/system-cards. Chapter 15.